Disconnection under load

I have dgraph running behind nginx wiith 3 alpha and one zero, the machines are 4 cores with 16G RAM and 160Gb ssd storage. With them being behind NGINX

My work load involved inserting a large amount of data. I found the most efficient way of inserting the data is to insert lots of mini relationships of new nodes and then later come back and merge them into the main graph blocks of nodes at a time. Then delete the nodes that weren’t needed.

If I run the the merge job while the insert is running there is a good chance I start to get rpc error: code = Unavailable desc = Bad Gateway: HTTP status code 502; transport: being reported by the clients.

There is nothing obvious in any of the logs, some stuff about Raft taking too long in the zero log. but that occurs every now and again if I am doing the merge or not. Is it worth running the zero on a separate machine? Could I be using all the resources up doing the work on the alpha node that shares the same machine as the zero?



This config is recommended for testing and studying, not a heavy load.

Check the prod checklist here https://dgraph.io/docs/deploy/production-checklist/#cpu-and-memory

I will check it out, but surely it should just be really slow if you have to little resources not fall over? I will look at upping the cores on the vms. Is it the alphas that don’t have enough cpu for raft or the zero? Would running the zero on another dedicated vm help?